Download the correlation and overlap dataset
The whole dataset as two CSV files, rebuilt after every US close from the same figures the API serves. No signup, no key, no rate limit.
| File | Rows | Size | Contents |
|---|---|---|---|
| pairs.csv | 52,937 | 1.7 MB | One row per tracked pair: correlations over 1, 3 and 5 years, annualized covariance, holdings overlap |
| symbols.csv | 4,756 | 0.3 MB | One row per asset: beta against SPY, annualized volatility, trailing returns, sector or fund group |
Data as of 2026-08-27. Both files are regenerated after the US close on every trading day, so a download is never more than one session behind the site.
Columns
pairs.csv holds asset_a and asset_b (tickers, alphabetical), corr_1y, corr_3y and corr_5y (Pearson correlation of weekly returns over the matching window), covariance_ann_pct2_3y (annualized covariance in percent squared) and holdings_overlap_pct (sum of minimum common weights, empty when either side does not publish holdings).
symbols.csv holds symbol, name, kind (stock or etf), group (sector for stocks, category for funds), beta_vs_spy_3y, volatility_ann_pct_3y and the trailing returns return_1y_pct, return_3y_pct and return_5y_pct, all in percent.
Load it
import pandas as pd pairs = pd.read_csv("https://www.pairbook.io/data/pairs.csv") diversifiers = pairs[pairs["corr_3y"] < 0.2].sort_values("corr_3y")
Every figure can be reproduced from public inputs with the formulas on the methodology page. For a single pair or a live lookup, the JSON API answers without a download.
Why download from here
Copies of this dataset exist on several archives, and they serve a different purpose: an archived snapshot is frozen on the day it was deposited, which is exactly what a paper needs to stay reproducible. The files on this page are the working copy. They are regenerated with the site, so a correlation you read here reflects last night's close rather than the state of the world on the day a snapshot happened to be uploaded.
If you only need part of the universe, the browsable views cost nothing to read: every asset has a hub under all stocks and all ETFs, sectors are grouped under sectors, and the extremes of the dataset are collected in the records.
License and citation
Both files are published under CC BY 4.0: use them freely, including commercially, as long as PairBook is credited with a link. An archived, versioned copy carries a DOI, which is the reference to use in a paper:
VoidLab. PairBook US Stock & ETF Correlation Snapshot. Zenodo. https://doi.org/10.5281/zenodo.22124847
The same snapshot is mirrored on Zenodo, Kaggle and Hugging Face, and the pipeline that produces it is open source. Figures come from end-of-day prices and issuer portfolio disclosures. Nothing here is investment advice.
